Desenvolvimento de plataforma de backtesting para testes de modelo de alta frequência
Carregando...
Orientador
Co-orientadores
Citações na Scopus
Tipo de documento
Trabalho de Conclusão de Curso
Data
2024
Título da Revista
ISSN da Revista
Título do Volume
Resumo
Este projeto visa o desenvolvimento de uma plataforma de backtesting para alta
frequência destinada à avaliação de modelos para a previsão de retornos em contratos futuros.
A plataforma será utilizada pelos responsáveis da área Quant da Legacy Capital. Para cumprir
com objetivo, foram utilizados raw market data de ordens e negociações realizadas na Bolsa
de Mercadorias & Futuros (B3), onde após manipulados e realizados os devidos cálculos,
servem como entrada para os modelos. Os modelos escolhidos foram Autorregressivo Exógena
(ARX), que é uma extensão dos modelos de regressão linear básica utilizada para séries
temporais, Long Short Term Memory (LSTM), que se configura como uma arquitetura de Rede
Neural Recorrente (RNN), Multilayer Perceptrons (MLP), que está dentro das Redes Neurais
Artificiais (RNA), LSTM-MLP, uma junção da arquitetura de Redes Neurais Artificias (RNA)
com Rede Neural Recorrente (RNN) e finalmente o Gradient Boosting Regressor (XGB), que
funciona como um processo de árvores de decisão sequencial. A plataforma será projetada para
ser flexível, permitindo a incorporação de diversos modelos e para ajuste de parâmetros
conforme determinado necessário pelo contratante, com o objetivo de aprimorar as capacidades
de previsão e a eficácia dos resultados advindo da plataforma.
The project aims the development of a high frequency backtesting platform designed to evaluate models that predict futures contracts returns. The platform will be used by the Quant team at Legacy Capital. To fulfill this objective, raw market data from the Commodities & Futures Exchange (B3) were used, which, after manipulation and appropriate calculations, served as input for the models. Some of the selected models included Autoregressive Exogenous (ARX), which is an extension of the linear regression model employed for time series analysis. Additionally, was used the Long Short-Term Memory (LSTM), that is a Recurrent Neural Network (RNN), along with Multilayer Perceptron (MLP) models within Artificial Neural Networks (ANNs). Also, a combination of MLP with LSTM, referred to as LSTM-MLP, was considered. Finally, the Gradient Boosting Regressor (XGB) was used, working as a sequential decision tree process. The platform is designed to be flexible, allowing for the incorporation of various models and parameter adjustments as determined necessary by the contracting party, with the aim of enhancing prediction capabilities.
The project aims the development of a high frequency backtesting platform designed to evaluate models that predict futures contracts returns. The platform will be used by the Quant team at Legacy Capital. To fulfill this objective, raw market data from the Commodities & Futures Exchange (B3) were used, which, after manipulation and appropriate calculations, served as input for the models. Some of the selected models included Autoregressive Exogenous (ARX), which is an extension of the linear regression model employed for time series analysis. Additionally, was used the Long Short-Term Memory (LSTM), that is a Recurrent Neural Network (RNN), along with Multilayer Perceptron (MLP) models within Artificial Neural Networks (ANNs). Also, a combination of MLP with LSTM, referred to as LSTM-MLP, was considered. Finally, the Gradient Boosting Regressor (XGB) was used, working as a sequential decision tree process. The platform is designed to be flexible, allowing for the incorporation of various models and parameter adjustments as determined necessary by the contracting party, with the aim of enhancing prediction capabilities.
Palavras-chave
Titulo de periódico
URL da fonte
Título de Livro
URL na Scopus
Idioma
Português
URL permanente
Notas
Projeto realizado para a empresa Legacy Capital
Membros da banca
Área do Conhecimento CNPQ
ENGENHARIAS